Field of the Art
The disclosure relates to the field of image analysis, and more particularly to the field of platforms for crowdsourcing image analysis activities.
Discussion of the State of the Art
Image analysis has been an important field of technology at least since the period of World War 2, when extensive use of image analysis, photogrammetry, and related technologies was used in conjunction with aerial photography for intelligence and bombing damage assessment purposes (among others). However, the extent of the use of image analysis (particularly image analysis of remotely-sensed images), particularly for identifying or locating targets of interest, has always been limited by the need for highly-trained, specialized image analysts or interpreters. The need for specialized (and expensive) skills has limited the use of image analysis to a correspondingly limited range of applications (notably military, homeland defense, and law enforcement).
The market for image analysis has also historically been limited by the high cost of obtaining images to analyze. In the military arena, the benefits were sufficiently apparent that large numbers of military reconnaissance flights were made over regions of interest since World War 2. But the cost of such flights virtually totally excluded all commercial applications of image analysis. Starting in the 1970s with the Landsat satellite, this began to change as low resolution satellite images became publicly available. A series of new satellites has opened up progressively more applications as the resolution, spectral coverage, geographic coverage, and cost per image have all continuously improved; accordingly, a significant market in commercial remote sensing imagery has emerged. But even this market has been limited from achieving its full potential because of the still-present requirement for expensive, scarce image analysis talent. Some progress has been made in automated image analysis technologies, but for a vast range of current and potential applications, large scale image analysis (such as would be needed when analyzing satellite images of a large region) remains too expensive and too supply-constrained to use.
One common type of image analysis problem is the “search and locate” problem. In this problem, what is needed is to find and to precisely locate one or more targets of interest. For example, in search and rescue, it may be important to find a missing plane using satellite imagery. Another example is the finding and precise location of warships, tanks, or other military targets of interest. Less common but promising applications include such things as assessing hurricane damage by finding and locating damaged buildings and infrastructure, finding and locating potentially important archeological sites (for instance, by identifying possible ruins in deserts), and assessing the scope of a refugee problem by for example counting tents in an area of interest.
Recently, the notion of “crowdsourcing” (using very large numbers of people, each doing a small part of a large task, to accomplish large of complex tasks quickly at extremely low cost) has emerged, and a number of crowdsourcing platforms have been implemented. Some of these address topics of broad general interest (for example, WIKIPEDIA™), and some are more specialized (for example, GALAXYZOO™, where users are shown images of objects from the Hubble Space Telescope and asked to decide if the object shown is a galaxy and, if so, what kind of galaxy it is). Most crowdsourcing platforms to date rely on volunteers to perform the work, although some (such as Amazon's Mechanical Turk) are commercial in nature and pay for crowdsourced work. There have been two general approaches to managing crowdsourced work. In the first, a large, complex or repetitive task is broken up into many subtasks, with each subtask being given to a single worker; as workers complete the subtasks, the results are rolled up and the overall task is completed at low cost. Generally, various means are used to measure the quality or value of the tasks performed by each participant, so that over time a reputation or quality score can be assigned to each participant; in some cases, work is assigned based at least in part on these reputation or quality scores. The work distribution and quality measurement approach is used, for example, by Amazon's Mechanical Turk platform. A second common approach to crowdsourcing is to use an essentially democratic process to have a crowd decide a difficult question. The process is referred to as “democratic” because each participant simply votes on what the participants believes the answer to be (this is helpful for classification problems such as that described above for GALAXYZOO™).
While aspects of both of these problems are relevant to the broad search and locate application domain, neither of them is sufficient. Consider the refugee assessment problem just described. The work distribution approach can clearly be used to divide up the task for distribution to many participants (typically volunteers). Similarly, the democratic approach could be used by the platform to decide whether something is or is not a tent, based on the number of votes each classification of a specific object received. But neither of these dominant approaches is satisfactory, and the two together are not satisfactory either, for the search and locate problem. It is not enough to divide and conquer, because in searching an image for a specific object considerable ambiguity will be present, and if each image segment is only viewed by a single person, there would be a high likelihood of missed targets (and indeed of false positives). If multiple participants are shown the same image and a vote is taken to decide if a target of interest is present, the outcome is better, but even in this case there are problems. Consider again the refugee problem—if there are in fact ten tents in a given field of view (image segment), various participants might report anywhere from three to twelve tents in the segment. A simple average of these counts could be taken, but would likely be inaccurate. But the “search and locate” problem also requires that the location of each tent be identified (at least implicitly—it is not so important in this particular problem that the exact location of each tent is known, but it is important to use locations to resolve count ambiguities; in most search and locate problems, though, the location aspects is a key output).
What is needed in the art is a platform for crowdsourcing the search and locate class of problems that accurately translates a large amount of crowdsourced inputs into an estimate of the precise locations of a number of targets of interest.